[tree-ssa] variables being dragged out of their lexical scopes

[Diego Novillo]

Given this program:

foo ()
{
  int i, k;

  {
    int j;
    j = i + 1;
  }

  k = i + 1;
  return i;
}

the conversion into SSA form is currently converting it into:

foo ()
{
  int i;
  int k;


  # BLOCK 0 (a.c:6).  PRED: -1.  SUCC: -2.
  {
    int j;

    j_2 = i_1 + 1
  };
  k_3 = j_2;
  return i_1;
}


Notice how j_2 is now used outside of its original lexical scope.
Apparently, this is a problem because moving variables outside
their scope affects debuggability (and maybe it has other side
effects?  I don't know).

Jason suggested that we could create additional temporaries in
the gimplifier so that we always convert assignments 'VAR = EXPR'
into 'TEMP = EXPR; VAR = TEMP':

foo ()
{
  int i;
  int k;
  int T.1;


  # BLOCK 0 (a.c:6).  PRED: -1.  SUCC: -2.
  {
    int j;

    T.1_5 = i_1 + 1;
    j_2 = T.1_5;
  };
  k_3 = T.1_5;
  return i_1;
}


Which works, but as Andrew pointed out, it could later be munged
again by copy-prop into:

foo ()
{
  int i;
  int k;
  int T.1;


  # BLOCK 0 (a.c:6).  PRED: -1.  SUCC: -2.
  {
    int j;

    T.1_5 = i_1 + 1;
    j_2 = T.1_5;
  };
  k_3 = j_2;
  return i_1;
}

I see two ways of handling this:

1- Never allow copy propagation in expressions of the form 'VAR = TEMP'
   (where 'VAR' is a program variable and 'TEMP' a compiler
   temporary).  This could work but I'm not sure if it would be a
   valid long term approach.

2- While building the flowgraph, emit a group-clobber instruction
   for all the variables that were declared inside a lexical
   scope.  The clobber instruction would be a new GIMPLE
   statement of the form CLOBBER_EXPR <var1, var2, ..., varN>.
   This would avoid any code movement that tries to take
   variables outside of their original scope (and it would also
   avoid having to generate all those temporaries in the
   gimplifier).


Thoughts?


Diego.